A Novel Approach to Control of Multilevel Converter Using Wavelets Transform
A. Jan Iwaszkiewicz1, B. Jacek Perz1
1
The Electrotechnical Institute, Gdansk Branch, Poland
80-557 Gdansk, Narwicka 1, Poland
phone:+48 58 3431291, fax+48 58 3431295, e-mail: A.jan.iwaszkiewicz@iel.gda.pl, B.jacek.perz@iel.gda.pl
Abstract
The paper presents an approach to the useful theory concerning
the problem of generating high quality AC voltage and current
waveforms in multilevel converters. The objective is to describe
a novel proposal of AC waveforms synthesis based on applying
analytical methods using sets of orthogonal wavelets. The
discussion includes a short survey of Haar wavelet and a pro-
posal of output waveforms synthesis based on wavelet trans-
form. The comparison of the proposed method with standard
solution is discussed. It has been done by use of such criteria as:
average square error and total harmonic distortion (THD).
Key words
Multilevel converter, Haar wavelet, wavelet transform,
wavelet waveforms synthesis.
1. Introduction
In contemporary industry and public area a great de-
mand has been appeared for higher power devices con-
verting electric energy. The devices like voltage and
current converters which are able to control and to supply
diverse equipment working in power range of hundreds
kW and more. These devices have to fulfil definite and
diversified requirements what implies diversified pur-
poses and methods of electric energy conversion. For
instance in AC drives where the possibility of speed
regulation is essential, the converter e.g. a voltage or
current source inverter (VSI or CSI) has to assure the
adjustment of output waveform fundamental level and
frequency[1, 2].
There are many other industrial applications e.g. un-
interruptible power suppliers (UPS) or distributed power
generation systems, where the essential demand is to
generate 50 or 60 Hz sinusoidal voltage waveforms. The
quality of generated waveforms, especially the Total
Harmonic Distortion factor (THD), should comply with
appropriate standards. Even if there is a need of output
waveform adjustment, the output parameters have to be
adjusted in a relatively small range.
In many devices like UPSs, active filters or voltage
regulators in electric energy grids the main and most
important features are: high quality of output waveforms,
output stability and efficiency of the device [4, 5, 6].
Similar requirements are to be fulfilled in converters
applied in renewable energy systems. In wind farms the
conversion of produced energy is inevitable in order to
solve the problem of unbalanced voltage levels and fre-
quency synchronization of many separated sources. The
similar problem appears in static DC sources like photo-
voltaic farms and fuel cells. The resulting DC power
should be converted to AC form using power electronics
converters.
During the first period of power electronics evolution
a two-level inverter was considered as the most suitable
device used to energy conversion in aforesaid applica-
tions. The features and performance as well as drawbacks
and limitations of the two-level inverter have been
largely recognized and verified in practice. Latest
achievements in power semiconductor technology permit
to work with higher switching frequency, but fast switch-
ing accompanying the PWM control causes power losses
in switching elements thus cutting inverter efficiency
down [7].
The range of the output power is a very important
and evident limitation of two-level inverters. The total
power is determined by the semiconductor switches
properties. This essential disadvantage can be reduced
using multilevel inverters. Recently, the multilevel in-
verters have emerged as a new and very important class
of converters. Thanks to their promising performance,
multilevel inverters are becoming more and more an
alternative to conventional two-level inverters. They
permit to overcome the problem of limited power and to
shape output waveforms. A result is that many multilevel
converters have been applied in industry.
The development of multilevel converters comprises
an area of research for new topologies, control strategies
and theory. Producing the required voltage or current
waveforms is possible in many ways: e.g. sinusoidal
PWM, selective harmonic elimination, space-vector
modulation (SVM) or shaping the stepped voltage or
current [8, 9, 10]. Important works and studies concern
the subject of frequency adjustment. Diverse methods of
converters’ control such as computing the adequate
switching angles of stepped waveforms or cancelling
specified harmonics have been developed [11]. In all
solutions a large number of semiconductor switches is
needed. The switches are not charged equally and their
admissible charge parameters are not fully utilized. It
results in decreasing of converter efficiency which is a
serious disadvantage especially in higher power applica-
tions.
The aforementioned disadvantages can be slightly
reduced by use of novel converters topologies as well as
mathematical tools aided control strategies. The paper
deals with technique of shaping the stepped output wave-
forms in multilevel converters. A mathematical approach
to the control strategy based on wavelets transform is
presented. The output waveform synthesis is accom-
plished using set of orthogonal wavelets. The discussion
includes such mathematical tools like Haar wavelet trans-
form. The comparison of methods and their impact on The equation defines an orthonormal wavelet family
topology is discussed in chapter 4. The comparative with dyadic scaling.
analysis has been done by use of three criteria: number of
DC sources needed, average square error and THD fac-
tor.

2. Haar wavelet function

The wavelets is a term for mathematical functions,
which allow the analysis of signals in different time scale
and with different resolution. The term „wavelets” is the
direct translation of French term „ondelettes”, which
means „little waves”. Thanks to the adjustable „scope of
the view” the wavelets can be used to distinguish and
analyse small and big details of the investigated process.
Especially they are useful in analyse of discontinuous or
stepped processes. The wavelets have been applied in
many not directly related areas like seismology, video
analysis, quantum mechanics or electronics [12, 13, 14,
15, 16, 17].
Till now the wavelets have been used mainly for Fig. 1. The scaling functions and and the fundamental Haar
analysis of processes or signals based on decomposition wavelet.
of the elements of the processes. The following consid- For given m the wavelet scale is 2m. The scale factor
erations will prove that wavelets can be also useful in m – settles the width and amplitude of the wavelet, and
composition of the power electronics signals and struc- the displacement factor n – settles the wavelet position on
tures. the time axis.
For this purpose the Haar wavelets have been
The fundamental Haar wavelet corresponds to the
adapted. The fundamental Haar wavelet can be con-
factors: m = 0 and n = 0 and can be denoted as
structed by transforming the following scaling function
ψ(t) = ψ00(t).
φ(t):

⎧1 for 0 ≤ t < 1, ⎫ 3. Wavelet waveforms synthesis

φ (t ) = ⎨ ⎬
⎩0 for other t ⎭
The Haar wavelet form is similar to the form of the
The composition of two consecutive scaling func- voltage or current pulse that can be obtained using simple
tions: one-phase inverter i.e. H-bridge cell. The displacement
and width of the wavelet can be freely controlled. Thanks
⎧1 for 0 ≤ t < 0,5),⎫
φ(2t ) = ⎨ ⎬
to these properties it is possible to apply wavelets in
⎩0 for other t ⎭ power electronics e.g. to form the output stepped wave-
forms of multilevel converters.
and
Let us define the scaling function φ(x) in an interval
⎧1 for 0,5 ≤ t < 1),⎫ x∈<0,2π):
φ(2t − 1) = ⎨ ⎬
⎩0 for other t ⎭ ⎧ 1 for 0 ≤ x < 2π, ⎫
φ (x ) = ⎨ ⎬
creates the fundamental Haar wavelet: ⎩ 0 for other x ⎭
ψ(t)=ϕ(2t) − ϕ(2t − 1) which can be defined as follows: The fundamental proposed wavelet is defined as the
Haar one alike:
⎧ 1 for 0 ≤ t < 0.5 ⎫
⎪ ⎪ ⎧ 1 for 0 ≤ x < π, ⎫
⎪ 1 ⎪
ψ (t ) = ⎨ − 1 for ≤ t < 1, ⎬ ⎪ ⎪
⎪ 2 ⎪ ψ ( x ) = ⎨ − 1 for π ≤ x < 2π, ⎬
⎪⎩ 0 for other t ⎪⎭ ⎪ 0 for other x ⎪
⎩ ⎭
The scaling functions and the fundamental Haar The wavelet determines one period of the rectangular
wavelet are presented in Fig. 1. wave and is mother function introducing a family of
Introducing two parameters: m – scale factor and n – wavelets:
displacement factor it is possible to write the family of ψ mn (x ) = ψ (2 − m x − 2πn ) for m, n = ..., − 2, − 1, 0, 1, 2, ...
wavelets: The wavelet scale is done as 2m2π and its displace-
ment on the x axis is determined as n-times 2m+1π. The m
ψ mn( t ) =
1
m
(
ψ 2−m t − n ) for m, n = ..., − 2, − 1, 0, 1, 2, ... factor scales not only the wavelet but the amplitude too.
2
 The scaling function and a few wavelets have been Denoting
presented in Fig. 2.
f mn (x ) = amn ψ mn (x )
it is possible to write fΨ ( x ) as a sum of components
fmn ( x ):
m =0 2− m −1
f ψ (x ) = ∑ ∑ f (x ) mn
m =−3 n=0

The components fmn ( x ) present component wave-

Fig. 2. The scaling function φ(x) and wavelets ψmn(x):
a) scaling function φ(x), b) fundamental wavelet ψ00(x),
c) wavelets ψ-1-1(x), ψ-20(x),ψ-22(x),ψ-37(x).
All wavelets ψmn(x) are orthogonal in the interval
x∈<0,2π). The defined statement creates a family of
orthogonal functions and can determine a basis of wave-
let transform. A continuous wavelet transform is defined
as:
∞
Wf (m, n ) = ∫ f (x ) ψ mn (x ) dx
−∞
lets, amplitude and phase of which is determined by
coefficients a mn.
Practically in power electronics applications, the ap-
proximation of a sine wave should be realized using a
finite number of wavelets. Natural aspiration of designers
is to utilize the possibly lowest number of components.
The accuracy of approximation depends on it. In mathe-
matics the accuracy is determined as an average square
error δ, a very useful criterion destined to that purpose. In
power electronics the most important criterion of the
approximated waveforms is THD factor.
Let us denote by fΨ ( x ) a waveform approximating
the function f (x)=sin(x) in the interval x ∈ < 0, 2 π ).
Assuming that fΨ ( x ) forms a combination of wavelets
determined by index m = -3, -2, -1, 0 it can be written as
a sum:
m=0 2− m −1 m=0 2− m −1
fψ (x) = ∑ amn ψmn (x) = ∑ ∑ f (x)

and presents itself a scalar product of a function f(x) and ∑

m=−3 n=0 m=−3 n=0
mn

function ψmn(x). A complete reconstruction of the func-

tion f(x) takes place when the inverse wavelet transform The coefficients a mn have been collected in Table 1.
is applied
TABLE 1: The wavelet coefficients a mn.
∞ ∞
f (x ) = C ∫ ∫ Wf (m, n ) ψ mn ( x )dm dn a mn n=0 n=1 n=2 n=3 n=4 n=5 n=6 n=7
−∞ −∞

The symbol C denotes a constant which can be calculated a 0n 0,6366 – – – – – – –

from Fourier transform of the function ψmn(x). a– 1n 0 0 – – – – – –

For applications more comfortable is to use a discrete a– 2n - 0,2637 0,2637 0,2637 - 0,2637 – – – –
inverse wavelet transform which is defined by the follow-
a– 3n - 0,1791 - 0,0742 0,0742 0,1791 0,1791 0,0742 - 0,0742 - 0,1791
ing equation:
∞ ∞
f (x ) = C ∑ ∑Wf (m, n ) ψ mn ( x) Successive steps of wavelet approximation fΨk ( x )
m = −∞ n = −∞ for k = 1, 2, 3 have been presented in Fig. 3 - 5.
It could be written as a sum of wavelets ψmn(x) multiplied First step of reconstruction creates function fΨ1 ( x )
by coefficients amn as a set of wavelets
∞ ∞ m =0 2− m −1

f (x ) = ∑ ∑a mn ψ mn ( x ) f ψ1 ( x ) = ∑ ∑ f (x ) = f mn −20 + f −21 + f −22 + f −23 + f −10 + f −11 + f 00

m =−2 n=0
m = −∞ n = −∞
in which two component wavelets are equal to zero ac-
The coefficients amn, called wavelet coefficients, are
cording to the Table 1. The resulting waveforms are pre-
scalar products of the function f(x) and wavelets ψmn(x).
sented in Fig. 3.
In the interval x∈<0,2π) they are given as
2π
Fig. 4 presents the result of the second step of ap-
proximation in which a few (but not all) wavelets f– 3n
amn = Cm ∫ f ( x )ψ mn ( x ) dx
0
have been added to the function fΨ1 ( x ). Function fΨ2( x )
creates a composition of following wavelets
The constant Cm is only dependent on coefficient m and
has the same value for different n. fψ2 ( x ) = f00 + f−20 + f−21 + f−22 + f−23 + f−30 + f−33 + f−34 + f−37
 Finally taking all wavelets for scale coefficients:
m = 0, m = – 2, m = – 3, the approximating function pre-
sents itself as a set of wavelets:

fψ3 ( x) = f00 + f−20 + f−21 + ...+ f−23 + f−30 + f−31 + ...+ f−36 + f−37

The resulting waveform is presented in Fig. 5.

Fig. 5. Third step of wavelet approximation:

fψ 3 ( x ) = fψ 2 + f −31 + f −32 + f −35 + f −36 .

Three approximating functions fΨ1 ( x ), fΨ2 ( x ),

fΨ3 ( x ) and related spectra have been presented below in
Fig. 3. First step of wavelet approximation: Fig. 6 - 11. The adequate values of THD factors and
average square errors δ have been also included.
( )
f ψ x = f −20 + f −21 + f −22 + f −23 + f 00 .
1

Fig. 6. Waveform of function fΨ 1 ( x )

Fig. 4. Second step of wavelet approximation: Fig. 7. Spectrum of fΨ 1 ( x ): THD = 23,06 %; δ=0,0252
( )
f ψ x = fψ 1 + f −30 + f −33 + f −34 + f −37 .
2
 rent sources. The angle α denotes the length of a shortest
step of the waveform and is equal to a half scale of the
shortest wavelet applied for approximation. Function fψ0
signifies one period of a rectangular wave which can be
treated as the lowest step of approximation.
TABLE 2: Fundamental parameters of approximating functions
fΨ 0 , fΨ 1 , fΨ 2 , fΨ 3 , fΨ 4 .

fΨk α N f ψk δψk THD

fΨ0 π 1 0, 0947 48, 37 %

Fig. 8. Waveform of function fΨ 2 ( x ). fΨ1 π /4 2 0, 0252 23, 0 6 %
fΨ2 π /8 3 0, 0091 13, 70 %
fΨ3 π /8 4 0, 0065 11, 44 %
fΨ4 π /16 8 0, 0016 5, 73 %

Fig. 9. Spectrum of fΨ 2 ( x ): THD = 13,70 %; δ=0,0091.
Fig. 10.Waveform of function fΨ 3 ( x ).
4. Comparative analysis
Data gathered in Table 2 summarize the approach to
wavelet synthesis - the expansion of the function
f(x)=sin(x) by use of wavelet series. It permits to com-
pare the features of the wavelet expansion with a typi-
cally used form of stepped waveforms in multilevel in-
verters. A cascade multilevel inverter where cells are
equal each other, will determine a reference object [8].
The DC link voltage is divided by number of cells in one
phase leg of the inverter and each cell is active during the
same interval of time. So it is assumed here that wave-
forms generated by such an inverter are composed of
steps of the same voltage and the same length α. The
angle length α of steps is determined by a number N of
steps in one period of the output waveform denoted as
FN. A comparison of parameters permitting to confront
classic and wavelet waveforms is presented in Table 3.
Data concerning the standard model are denoted by
index N while data of the wavelet model - by index ψ.
Remaining symbols have the following meaning:
− α min – the shortest step angle,
− NDC – number of demanded DC sources,
− δ – approximation error,
− THD – total harmonic distortion factor.
TABLE 3: Comparative parameters of two models of AC wave-
form approximation.

fΨk ∪ FN α min NDC δ THD

fΨk FN fΨk FN N fψ k N FN δψk δN fΨk FN

fΨ0 FN = 2 π π 1 1 0,0947 0,0947 48,37% 4 8 , 3 7 %

Fig. 11. Spectrum of fΨ 3 ( x ): THD = 11,44 %; δ=0,0065.
fΨ1 FN = 6 π/4 π/3 2 2 0,0252 0,0440 23,60% 3 1 , 0 9 %
Selected parameters describing features of the wave-
fΨ2 F N = 1 2 π/8 π/6 3 3 0,0091 0,0113 13,70% 1 5 , 2 3 %
let applications in shaping stepped waveforms have been
collected in Table 2. Besides average square errors δ and fΨ3 F N = 1 6 π/8 π/8 4 4 0,0065 0,0064 11,44% 1 1 , 4 1 %
THD factors a N f parameter has been included. It fΨ4 F N = 2 4 π/16 π/12 8 6 0,0016 0,0028 5 , 7 3 % 7 , 6 3 %
ψk

denotes a number of different absolute values in one

period of the stepped waveform. This parameter relates The comparison of parameters confirms that for the
directly to the number of demanded DC voltage or cur- same number of independent sources the THD factor
varies depending on the used model. Wavelet synthesis
has advantages in relation to the standard solution when
considering on lower levels of approximation.
On the second step of expansion the wavelet model
produces shorter angles α min : α = π / 4 related to π / 3 in
the waveform FN = 6. and generates waveforms with sig-
nificantly reduced δ and THD. In both models the con-
verter needs two DC sources but in wavelet converters
the Dc voltages are different. Thanks to dyadic scaling in
wavelet transform definition (8) it results in elimination
of third and fifth harmonic in fψ1 spectrum. The THD
value 23, 06 % reaches almost the lowest level available
in three level inverters. In Fig. 12 and 13 two spectra of
considered waveforms are presented: spectrum of the
waveform generated in a standard three level inverter and
spectrum of the waveform obtained in the wavelet con-
verter.
Fig. 12. Spectrum of fN(x) for N=6: THD=31,09%; δ=0,0440.
Fig. 13. Spectrum of: fψ1(x): THD=23,06 %; δ=0,0252.
The next step of approximation permits to obtain
also better waveforms in wavelet converter than in stan-
dard one. Taking as a criterion the aforementioned pa-
rameters α min , δ and THD factor it is possible to compare
adequately: π / 8 versus π /6, 0, 0091 to 0, 0113 and
13, 70 % to 15,23 %.
Following steps of wavelet approximation bring
similar results as received in case of considered standard
stepped waveforms. However it seems that the wavelet
converter presents better possibilities with regard to con-
trol of fundamental voltage and frequency of output
waveforms because it is equipped in diversified DC
sources and can operate with shorter pulses. Although the
problem is very important in AC drives control but is not
considered in this paper.
5. Wavelet converter
One-phase wavelet voltage converter, generating fψ1
waveform, consists of two voltage inverters (respectively
transistors T1÷T4 and T5÷T8) connected in cascade. In
each phase of the complex converter the component in-
verters are supplied from two independent voltage
sources: UD1 and UD2. The supply voltages are propor-
tional to the relevant amplitudes of component wavelets.
For the converter output voltage of 500 V the voltage
UD1=318 V, and UD2=132 V. The schematic diagram of
three-level three-phase converter, consisting of three one-
phase component inverters is presented in Fig. 14.
Fig. 14. Three-phase three-level converter as an example of
wavelet theory application.
The control of such a converter is derived from de-
scribed wavelets model. The inverter consisting of tran-
sistors T5÷T8 generates waveform UF1, corresponding
with wavelet f00, in each period of output voltage. It
means that transistors T5 and T8 are switched on in the
first half of the period while transistors T6 and T7 are
switched on in the second one. The inverter consisting of
transistors T1÷T4 generates four waveform UF2, corre-
sponding with wavelets f-2n, in each period of output
voltage. Because the wavelets in the middle of the output
waveform have a different phase shift than the other
ones, the periodical switching pattern is as following: T2-
T3, T1-T4; T1-T4, T2-T3; T1-T4, T2-T3; T2-T3, T1-T4.
The summing process of the wavelets in cascade is
accomplished by serial connection of the outputs of com-
ponent inverters. When the transformer is used as a
summing node the frequency corresponding with wave-
lets f-2n is 200 Hz. Anyhow due to the phase change there
will appear pulses of 5 ms duration two times in each
output voltage period.
It is worth mentioning, that the power rating of the
component inverters is not equal. The power of the in-
verter generating the wavelets f-2n (transistors T1÷T4) is
much lower than the power of the second one generating
wavelet f00 (transistors T5÷T8).
The topology of multilevel converter is dependent on
the method used for its output waveform synthesis. The
proposed method based on wavelet transform is very
suitable for cascade converters even so it demands inde-
pendent voltages sources.
5. Conclusions
Advantageous features of wavelets permit to use
them not only for analysis of processes consisting in
decomposition but on the contrary in composition of
signals. The wavelets may be also applied in composition
of signals and structures in power electronics. The paper
presents a theory of waveform synthesis based on wave-
lets transform which may be a useful mathematical tool
in designing of multilevel converter structures and con-
trol strategies.
The wavelet transform synthesis applied to control
multilevel converters has good features in relation to
standard methods. The wavelet converter becomes auto-
matically an asymmetric multilevel inverter because it
assures better and faster adaptation of output waveforms
to sine wave shape. For low level of approximation in-
cluding three and four level inverters, the wavelet synthe-
sis proves this feature and permits to obtain better ap-
proximation of waveforms. Furthermore the wavelet
control method has the following features:
− low harmonic content of output waveforms,
− relatively low switching frequency of all inverters,
− diversified power rating of component inverters,
− simple converter control circuit.
The mathematical wavelet model of converter is
based on discrete inverse wavelet transform. Thanks to
its universality the model gives an impact for develop-
ment of converter topologies and control strategies. It
may be developed by incorporating suitable control func-
tions e.g. PWM thus permitting to be applied in AC
drives.
The possible application area belongs mainly to
higher power converters especially in high power unin-
terruptible power supplies, voltage regulators like
FACTS and UPQC, distributed power generation systems
with DC sources like photovoltaic or fuel cells with dis-
crete voltage levels. In the sources of this kind there is an
easy way to group cells to obtain the desired voltage
levels for the multilevel converter.
Acknowledgement
The scientific work was supported by The Polish
Ministry of Science and Higher Education in years
2006 - 2009 as a research project.
References
[1] Akagi H.: Control Strategy and Sections of a Shunt Ac-
tive Filter for Damping of Harmonic Propagation in
Power Distribution Systems. IEEE TRAN. on Power
Delivery, January 1997.
[2] Blaabjerg F., Jaeger U., Munk-Nielsen S., Pedersen J. K.:
Power Losses in PWM_VSI Inverter Using NPT or PT
IGBT Devices. IEEE Transactions on Power Electronics,
vol. 10, no. 3, May 1995.
[3] Białasiewicz J. T.: Falki i aproksymacje. Wydawnictwa
Naukowo-Techniczne, Wydanie drugie, Warszawa 2004.
[4] Chiasson J. N., Tolbert L. M., McKenzie K. J., Zhong
Du.: Control of a Multilevel Using Resultant Theory.
IEEE Transactions on Control Systems Technology, vol.
11, no. 3, May 2003.
[5] Daubechies I.: The wavelet Transform, time-frequency
localization and signal analysis. IEEE Transactions on
Informatics Theory, 1990, vol. 36, s. 961-1005.
[6] Faranda R., Valade I.: UPQC Compensation Strategy and
Design Aimed at Reducing Losses. IEEE International
Symposium on Industrial Electronics ISIE 2002, vol. 4.
[7] Graps A.: An Introduction to Wavelets, IEEE Computa-
tional Science and Engineering, summer 1995, vol. 2, no.
2.
[8] Haar A.: Zur Theorie der orthogonalen Funktionensys-
teme, Mathematische Annalen, 1910, Vol. 69, pp. 331-
371.
[9] Hashad M., Iwaszkiewicz J.: A Novel Orthogonal-
Vectors-Based Topology of Multilevel Inverters, IEEE
Transactions on Industrial Applications, vol. 49, no. 4,
August 2002.
[10] Iwaszkiewicz J., Hashad M.: Convertor Control with
Help of Orthogonal Space Vectors – Novel Control Strat-
egy and Novel Topology, PCC Osaka 2002.
[11] Iwaszkiewicz J.: A Novel Recurrent Approach to the
Output Voltage Filtering of the Inverter, EPQU 2003.
[12] Iwaszkiewicz J., Perz J.: Amplitude Modulation – an
Alternative Method of Generating the Convertor
Output Waveforms, ICREPQ’04, Barcelona 2004.
[13] Iwaszkiewicz J., Perz J. – „Multilevel Convertors for
Distributed Power Generation Systems with DC Voltage
Sources”, International Conference on Renewable Energy
and Power Quality ICREPQ’05, Saragossa 2005.
[14] Iwaszkiewicz J., Perz J. – „Mathematical Approach to
THD Optimisation in Multilevel Convertors”, Interna-
tional Conference on Renewable Energy and Power
Quality ICREPQ’06, Palma de Mallorca 2006.
[15] Iwaszkiewicz J., Perz J. – „A Novel Approach to Sine-
wave Approximation in Multilevel Converters”, The
IEEE Industry Applications Society Workshop on Re-
newable Energy Based Units and Systems (REBUS), St.
Petersburg 2006.
[16] Iwaya K., Takahashi I.: Novel Multilevel PWM Wave
Control Method using Series Connected Full Bridge
Inverters, Proceedings of the IEEE Electric Machines and
Drives Conference, volume 3, pages 1543-1548, June
2003.
[17] Kawabata T., Kawabata Y., Nishiyama K.: New Configu-
ration of High-Power Inverter Drives," Proceedings of the
IEEE International Symposium on Industrial Electronics,
volume 2 , pages 850 -855, June 1996.
[18] Kaźmierkowski M., P., F. Blaabjerg F.: Impact of Emerg-
ing Technologies on PWM Control of Power Electronic s,
IEEE Industrial Electronics Society Newsletter, Decem-
ber, 1995.
[19] Lida S., Masukawa S., and Kubota Y.: Improved Voltage
Source Inverter With 18-Step Output Waveforms, IEEE
Industry Applications Magazine, January/February 1998.
[20] Rademacher H.: Einige Satze von allgemeinen orthogo-
nalfunktionen. Mathematische Annalen, 1922, pp. 122-
138.
[21] Rodriguez J., Lai J.-S., Peng F.Z..: Multilevel Inverters:
A Survey of Topologies, Controls and Applications, IEEE
Transactions on Industrial Electronics, vol. 49, no. 4, pp.
724-738, August 2002.
[22] Sten K., Stacey E.: UPFC-Unified Power Flow Control-
ler: Theory, Modeling, and Applications, IEEE Transac-
tions on Power Delivery. Vol. 13. No. 4. October 1998.
[23] Walsh J. L.: A closed set of orthonormal functions. Amer.
Journal of Mathematics, 1923, no. 1 pp. 5-24.